Degree Type

Dissertation

Date of Award

2012

Degree Name

Doctor of Philosophy

Department

Agronomy

First Advisor

Jode Edwards

Second Advisor

Kendall Lamkey

Abstract

Yield trials data are essential to the process as they provide information regarding the performance of the genetic material being tested across multiple locations and seasons. The analysis of the data is complicated by the presence of genotype-by-environment interaction (GEI). A Bayesian approach was used to estimate variance components in a hierarchical model that allows for heterogeneous error and GEI variances applied to corn yield data from the Iowa Crop Performance Test carried out between 1995 and 2005. An average of 508 hybrids per year was tested with very little overlap between locations and years, which resulted in a very unbalanced data set. We divided the data into 16 subsets to study the effect of variability across locations and across years. All sub sets presented strong evidence of heterogeneity at both GEI and error variance levels. With the information gathered from the first study, we designed a simulation study and generated data under 33 different conditions of repeatability, sample size and level of heterogeneity, including 6 data sets simulated from a model that had no heterogeneity (homogeneous model); we considered these sets our control. The objective of this part of our work was to determine if there are any set of conditions where (i) selections made under the heterogeneous model lead to bigger increments in yield in comparison with the homogeneous model (ii) the heterogeneous model provides"better" parameters, in terms of accuracy and precision (iii) the heterogeneous model leads to select more stable cultivars. The heterogeneous model was able to pick up the lack of variability of the data from the control sets, and performed very similarly to the homogeneous model suggesting a broader use of the model even when the heterogeneity is not detectable. In the case of data simulated from the heterogeneous model, the estimates of the component of variance where more accurate when using the heterogeneous model, we observed a very large improvement on the estimates of GEI variances, which in our study was the only term truly heterogeneous. The Bayesian estimator penalizes the more unstable genotypes bringing their average towards the overall mean, leading to selection of more stable cultivars. The genetic gain differential after selection using the heterogeneous model was small, but it consistently showed an advantage in favor of the heterogeneous model suggesting that selections made under the heterogeneous model will lead to bigger increments in yield in comparison with the homogeneous model.

DOI

https://doi.org/10.31274/etd-180810-1740

Copyright Owner

Massiel Andrea Orellana

Language

en

File Format

application/pdf

File Size

80 pages

Share

COinS